Integration of Vision Transformer Networks in YOLO for Object Detection: Comparative Study on Plant Disease Detection

This paper presents a comparative study on plant disease detection using Vision Transformer Networks (ViT) integrated with YOLOv26 and RT-DETR for object detection, aiming to improve the accuracy and efficiency of plant disease classification. We analyze the performance of these models in detecting a wide variety of plant diseases, including Beans Angular LeafSpot, Strawberry Leaf Spot, and Tomato Blight, among others. The models are evaluated based on their precision, recall, and mAP scores, with particular focus on detecting complex multi-class diseases in agricultural images. The ViT-based integration with YOLOv26 and RT-DETR demonstrates a significant improvement in detection accuracy over traditional object detection models, supported by advanced post-processing techniques like Soft-NMS and Grad-CAM for interpretability. The comparative analysis highlights the strengths and weaknesses of each model, providing insights into their applicability for real-time plant disease detection systems. Furthermore, the study explores the potential for deploying these models in agricultural environments to aid farmers in quick and reliable disease diagnosis.

Keywords: Plant Disease Detection, Vision Transformer, YOLOv26, RT-DETR, Soft-NMS, Grad-CAM, Object Detection, Deep Learning, Agricultural Imaging, Precision Agriculture